STOCHASTIC PROCESSES IN FREEWAY TRAFFIC PART II. INCIDENT DETECTION ALGORITHMS

Part I of this two-part series of papers addressed the problem of constructing stochastic process models for freeway traffic and examined the robustness of these models in providing short-term predictions of traffic state variables. In this part, an approach for the automatic detection of freeway traffic incidents (accidents, stalled vehicles, debris on the road, etc.) is presented based on the ARIMA (0,1,3) model described in Part I. Traffic incidents often result in full or partial lane blockage, and they occur most frequently during the peak periods when normal freeway capacity is already hard pressed. West found that the non-recurring congestion due to traffic incidents is responsible for as much motorist delay on urban freeways as is the recurring congestion due to geometric bottlenecks. Many freeway operating agencies, therefore have recognised the fact that incident management measures, if soundly based can permit greater traffic flows to be accommodated more safely on the existing freeway system with considerable savings in travel time to motorists.